Distributive Video Coding

1
Distributive Video Coding
By Nagaraja Shivashankar
2
Agenda

• Problem and Motivation.
• Slepian Wolf Coding.
• Wyner-Ziv Coding.
• Stanfords Distributive Video Coding.
• Berkeleys Distributive Video Coding.
• Comparison with Conventional codec.

3
Problems and Motivation

• Conventional (Hybrid) Video Codecs (CVC)
• Encoders are 5 to 10 times more complex than
  Decoders.
• Prone for the error drift.
• Well suited for Broadcasting, Video On Demand
  (one to many)
• Can these CVC be used in wireless Video sensors
  or Mobile camera which has
• Limited Processing capabilities
• Low power budget
• Information loss transmission loss or Error
  drift.
• Distributive Video Coding (DVC) is the
  consequence of information-theoretic bounds
  established in 1970s
• By Slepian and Wolf Wyner and Ziv
• The traditional balance of complex encoder and
  simple decoder is essentially reversed

4
SlepianWolf Theorem
DMSS (X,Y) p(x,y), For Joint Encoding R
H(X,Y) is sufficient. For Separately
Encoding R H(X)H(Y) is sufficient.
but Slepain Wolf showed that R H(X,Y) is still
sufficient for statistically correlated
sources Achievable region for distributive
source coding is given by Rx H(XY), Ry
H(YX), RxRy H(X,Y).
5
SlepianWolf Theorem Contd..
How do we code X with H(XY) given bits and
recover the information at the decoder?
6
SlepianWolf Coding Example

• If Y(side information) is available to both enc
  and dec
• X and Y equiprobable 3-bit binary words.
• Correlation Hamming dist, dH(X,Y) 1.
• H(X) 3 bits.
• X Y 000,001,010,100.
• H(XY) H(XYY) 2 bits.
• Total H(x,y) H(y)H(xy) 5bits
• Enc f(X, Y) f(X Y)
• Dec g(W, Y) g(W) Y
• What if Y is not present in Encoder ?

7
SWC Example Contd..

• From Slepain-Wolf theorem it is still possible
  to send H(xy) 2b instead of H(x) 3b for X
  without loss at the decoder.
• Partitioning X into Four bins or Cosets
• 000,111,100,011,010,101,001,110
• Encoder sends 2 bit index of coset or bin that X
  belongs.
• Decoder resolve the uncertainty by checking which
  is closer in hamming distance to Y and declaring
  that value of X.

8
Coding with side information

• A special case of the distributed coding problem
• Side information Y is available at the decoder
  but not at the encoder
• RY H(Y) is achievable for encoding Y
• RX H(XY) , regardless of the encoders access
  to side information Y

9
Wyner-Ziv Theorem

• Wyner and Ziv extended the work by Slepian and
  Wolf by studying the lossy case in the same
  scenario, where signals X and Y are statistically
  dependent.
• Y is transmitted at a rate equal to its entropy
  (Y Side Information) and what needs to be found
  is the minimum transmission rate for X that
  introduces no more than a certain distortion D.
• Wyner-Ziv coding suffers rate loss when compared
  to lossy coding of X when side information Y is
  available at both enc and decoder

10
Wyner-Ziv Codec

• In general there is a rate loss with two
  exceptions
• Quadratic Gaussian case.
• X YZ, Z is independently Gaussian but X and Y
  could be general distributions.
• Wyner-Ziv source-channel coding problem
• Quantization loss and binning loss.
• Wyner-Ziv limit Efficient source channel
  codes.
• Decoder Rely more on X and Y at high and low bit
  rates, respectively.

11
Distributed Video Coding Models

• In the Literature, there are essentially two
  research groups who have been responsible for the
  development of the most relevant distributed
  source video coding systems
• Dr. Kannan Ranachandran's Group at Berkeley
  -Berkeley Wyner-ziv Robust Video coding solution.
• Dr. Bernod Girod's groups at stanford -
• Stanford Wyner-ziv Low complexity Video coding
  solution.

12
Stanford Model

• The Wyner-Ziv frames W Corresponds to main
  information (Sequence X).
• The Information resulting from the motion-
  compensated extrapolation Module, W is the side
  information (sequence Y). In turn Yk refers to
  the side information.
• Reconstruct E(Xkqk,Yk)

13
Experimental results
14
Salesman at 10 fps
DCT-based Intracoding 149 kbps PSNRY30.0 dB
Wyner-Ziv DCT codec 152 kbps PSNRY35.6 dB
GOP8
15
Berkeley Model(PRISM)

• The main information corresponds to the quantized
  transform coefficients (sequence X).
• The side information is composed of candidates to
  prediction block. The candidates to prediction
  blocks are generated through half-pixel motion
  search in the previous reconstructed frame.

16
Experimental results

17
(No Transcript)
18
Conclusions

• The authors believe that distributed coding
  techniques will soon complement conventional
  video coding to provide the best overall system
  performance and enable novel applications.
• Examples of DVC systems include wireless
  videosensors for surveillance, wireless PC
  cameras,mobile camera phones, and networked
  camcorders. In all these cases compression must
  be implemented at the camera where memory and
  computation are scarce.
• Useful for wireless video applications by means
  of transcoding architecture use.

19
Thanks and Questions
20
SlepianWolf Coding Example

• 2. Let X Y be two correlated 8-b grayscale
  image, with x y being pixel locations. where x
  y-3,y-2 y4 or -3 lt x-y lt4 (8 different
  values).

21
(No Transcript)